How to Spot AI Scams: 7 Rules for Deepfakes, Fake Calls and AI Errors

So the evidence is strong for AI-assisted fraud, impersonation and misleading AI marketing. It is weaker for broad claims about the everyday error rate of AI systems; the sources used here do not provide a reliable general percentage for how often AI tools get things wrong.

1. Stop whenever pressure is part of the message

Urgency is one of the clearest warning signs. Be extra careful with any call, message, video or chatbot response that tells you to act immediately — pay now, click now, confirm now, keep this secret, or send sensitive details before you have time to think.

Use a fixed pause rule:

• Do not pay immediately.
• Do not open links under pressure.
• Do not download attachments.
• Do not approve a payment or account change.
• Do not share sensitive data.
• Verify through a separate route first.

That pause matters because generative AI can help scammers create more convincing false identities, fake media and malicious websites .

2. Verify people through a second channel you choose

A voice on the phone or a face on a video call is no longer enough. The FTC’s proposed protections against impersonation of individuals specifically refer to fraud risks from AI-generated deepfakes .

In practice, this means you should end the current contact and check through a channel you control. Call a number you already had saved. Type the official website into your browser yourself. Contact the person through a known, independent route — not through the link, number or account that just contacted you.

Useful habits include:

• a family code word for real emergencies;
• a standing callback rule for any money request;
• no exceptions for claims of urgency;
• no verification inside the same chat, call or video call.

If someone is legitimate, they can tolerate a careful check. If they are pressuring you not to check, that is the warning.

3. Never share codes, passwords or payment approvals in the live contact

One-time codes, passwords, ID photos, wallet access and payment approvals are high-risk data. If someone asks for them during an incoming call, chat, video call or voice message, keep the response boring and consistent: stop, disconnect and verify separately.

That rule applies even if the person appears to be from your bank, a government office, your employer, a platform support team, a customer, a family member or a manager. The CFTC specifically lists fake voices, videos, live-streaming video chats, social media profiles and malicious websites as tools fraudsters may use to make AI-assisted deception more credible .

4. Treat voices, videos and screenshots as clues — not proof

Do not wait until you can “spot the deepfake” with your eyes or ears. For most people, the safer test is not technical. It is practical: Can the story be confirmed another way?

Before acting, ask:

• Does this request fit the person’s or organisation’s usual behaviour?
• Is there a known callback route?
• Is secrecy being demanded?
• Are you being asked to send money to a new account, wallet or unusual payment method?
• Are bank details, delivery details or login details being changed?
• Would this still make sense if the voice, image or screenshot were fake?

If the answers feel off, stop the transaction. For money, identity and access, a second verification route should be mandatory.

5. Be sceptical of AI money promises

‘AI-powered’ is not due diligence. It does not prove that a product works, that an investment is safe or that an income claim is realistic.

The FTC announced action in 2024 against deceptive AI claims and schemes, including a lawsuit against an online business opportunity that allegedly claimed its “cutting edge” AI-powered tools would help consumers quickly earn thousands of dollars a month in passive income by opening online storefronts .

Be especially cautious with offers such as:

• AI trading systems with supposedly safe profits;
• passive-income schemes based on AI automation;
• guaranteed returns from an algorithm;
• expensive courses or tools built around unrealistic income claims;
• testimonials, screenshots or demo videos without verifiable evidence.

The safer rule: check who is behind the offer, what the contract says, what it costs, what the risks are and whether the primary evidence supports the claim. Do not pay simply because something is marketed as AI.

6. Use AI answers as drafts, not final decisions

The same principle that protects you from AI scams also protects you from AI mistakes: do not copy, forward or act on important AI output until you have checked it.

For high-stakes decisions, use this workflow:

1. Open the primary source. That may be a law, official agency page, original study, contract, bank statement, manufacturer documentation or medical information from a qualified source.
2. Check the numbers and dates. Look at the context, definitions, publisher and update date.
3. Get human review when the risk is high. This is especially important for legal, medical, financial, safety, identity and workplace decisions.
4. Do not forward AI summaries unchecked. Verify the central claims before you share them as fact.

AI can be useful for drafting, organising and summarising. It should not be the only reason you decide something is true, safe or legally reliable.

7. Set rules for family and work before there is a crisis

The best protection is a routine everyone already knows. That matters because impersonation fraud is a major harm area: the FTC says scams impersonating businesses and government agencies are consistently among the top reported frauds, with $2.95 billion in reported consumer losses in 2024 — again, not limited to AI cases .

For families:

• Agree on a code word for emergencies.
• Always call back for money requests, using a known number.
• Tell children, parents and grandparents that voices and videos can be faked.
• Never share one-time codes or passwords by phone.

For teams and businesses:

• Do not allow payment approvals based only on chat, email, voice message or video call.
• Double-check changes to bank details.
• Use a two-person review for unusual payments.
• Require callbacks through known numbers.
• Check key facts in AI-generated content before using it externally.

A five-step checklist for the moment something feels wrong

If a call, video, chat or AI-generated message pushes you to act, run through this checklist:

1. Pause. Do not react immediately.
2. Protect the data. Do not share codes, passwords, ID documents or payment approvals.
3. Switch channels. Verify through a known number or an official website you opened yourself.
4. Look for the source. Check the original account, contract, agency page, bank record or direct confirmation.
5. Bring in another person. For money, identity and major decisions, do not act alone under pressure.

Bottom line

The best everyday defence against AI scams and AI mistakes is not a special tool. It is a repeatable habit: stop, verify separately, protect sensitive data and check important AI claims against original sources.

The sources behind this article strongly support the risks around AI-assisted deception, deepfake impersonation, AI-generated calls and misleading AI claims . They do not establish a general error rate for everyday AI systems. The regulatory sources are also mostly from the United States, so local reporting routes and legal consequences may differ elsewhere.